xref: /freebsd/sys/netinet/udp_usrreq.c (revision 262e143bd46171a6415a5b28af260a5efa2a3db8)
1 /*-
2  * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  * 4. Neither the name of the University nor the names of its contributors
14  *    may be used to endorse or promote products derived from this software
15  *    without specific prior written permission.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  *
29  *	@(#)udp_usrreq.c	8.6 (Berkeley) 5/23/95
30  * $FreeBSD$
31  */
32 
33 #include "opt_ipsec.h"
34 #include "opt_inet6.h"
35 #include "opt_mac.h"
36 
37 #include <sys/param.h>
38 #include <sys/systm.h>
39 #include <sys/domain.h>
40 #include <sys/jail.h>
41 #include <sys/kernel.h>
42 #include <sys/lock.h>
43 #include <sys/mac.h>
44 #include <sys/malloc.h>
45 #include <sys/mbuf.h>
46 #include <sys/proc.h>
47 #include <sys/protosw.h>
48 #include <sys/signalvar.h>
49 #include <sys/socket.h>
50 #include <sys/socketvar.h>
51 #include <sys/sx.h>
52 #include <sys/sysctl.h>
53 #include <sys/syslog.h>
54 
55 #include <vm/uma.h>
56 
57 #include <net/if.h>
58 #include <net/route.h>
59 
60 #include <netinet/in.h>
61 #include <netinet/in_systm.h>
62 #include <netinet/in_pcb.h>
63 #include <netinet/in_var.h>
64 #include <netinet/ip.h>
65 #ifdef INET6
66 #include <netinet/ip6.h>
67 #endif
68 #include <netinet/ip_icmp.h>
69 #include <netinet/icmp_var.h>
70 #include <netinet/ip_var.h>
71 #include <netinet/ip_options.h>
72 #ifdef INET6
73 #include <netinet6/ip6_var.h>
74 #endif
75 #include <netinet/udp.h>
76 #include <netinet/udp_var.h>
77 
78 #ifdef FAST_IPSEC
79 #include <netipsec/ipsec.h>
80 #endif /*FAST_IPSEC*/
81 
82 #ifdef IPSEC
83 #include <netinet6/ipsec.h>
84 #endif /*IPSEC*/
85 
86 #include <machine/in_cksum.h>
87 
88 /*
89  * UDP protocol implementation.
90  * Per RFC 768, August, 1980.
91  */
92 #ifndef	COMPAT_42
93 static int	udpcksum = 1;
94 #else
95 static int	udpcksum = 0;		/* XXX */
96 #endif
97 SYSCTL_INT(_net_inet_udp, UDPCTL_CHECKSUM, checksum, CTLFLAG_RW,
98 		&udpcksum, 0, "");
99 
100 int	log_in_vain = 0;
101 SYSCTL_INT(_net_inet_udp, OID_AUTO, log_in_vain, CTLFLAG_RW,
102     &log_in_vain, 0, "Log all incoming UDP packets");
103 
104 static int	blackhole = 0;
105 SYSCTL_INT(_net_inet_udp, OID_AUTO, blackhole, CTLFLAG_RW,
106 	&blackhole, 0, "Do not send port unreachables for refused connects");
107 
108 static int	strict_mcast_mship = 0;
109 SYSCTL_INT(_net_inet_udp, OID_AUTO, strict_mcast_mship, CTLFLAG_RW,
110 	&strict_mcast_mship, 0, "Only send multicast to member sockets");
111 
112 struct	inpcbhead udb;		/* from udp_var.h */
113 #define	udb6	udb  /* for KAME src sync over BSD*'s */
114 struct	inpcbinfo udbinfo;
115 
116 #ifndef UDBHASHSIZE
117 #define UDBHASHSIZE 16
118 #endif
119 
120 struct	udpstat udpstat;	/* from udp_var.h */
121 SYSCTL_STRUCT(_net_inet_udp, UDPCTL_STATS, stats, CTLFLAG_RW,
122     &udpstat, udpstat, "UDP statistics (struct udpstat, netinet/udp_var.h)");
123 
124 static void udp_append(struct inpcb *last, struct ip *ip, struct mbuf *n,
125 		int off, struct sockaddr_in *udp_in);
126 
127 static int udp_detach(struct socket *so);
128 static	int udp_output(struct inpcb *, struct mbuf *, struct sockaddr *,
129 		struct mbuf *, struct thread *);
130 
131 void
132 udp_init()
133 {
134 	INP_INFO_LOCK_INIT(&udbinfo, "udp");
135 	LIST_INIT(&udb);
136 	udbinfo.listhead = &udb;
137 	udbinfo.hashbase = hashinit(UDBHASHSIZE, M_PCB, &udbinfo.hashmask);
138 	udbinfo.porthashbase = hashinit(UDBHASHSIZE, M_PCB,
139 					&udbinfo.porthashmask);
140 	udbinfo.ipi_zone = uma_zcreate("udpcb", sizeof(struct inpcb), NULL,
141 	    NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
142 	uma_zone_set_max(udbinfo.ipi_zone, maxsockets);
143 }
144 
145 void
146 udp_input(m, off)
147 	register struct mbuf *m;
148 	int off;
149 {
150 	int iphlen = off;
151 	register struct ip *ip;
152 	register struct udphdr *uh;
153 	register struct inpcb *inp;
154 	struct mbuf *opts = 0;
155 	int len;
156 	struct ip save_ip;
157 	struct sockaddr_in udp_in;
158 
159 	udpstat.udps_ipackets++;
160 
161 	/*
162 	 * Strip IP options, if any; should skip this,
163 	 * make available to user, and use on returned packets,
164 	 * but we don't yet have a way to check the checksum
165 	 * with options still present.
166 	 */
167 	if (iphlen > sizeof (struct ip)) {
168 		ip_stripoptions(m, (struct mbuf *)0);
169 		iphlen = sizeof(struct ip);
170 	}
171 
172 	/*
173 	 * Get IP and UDP header together in first mbuf.
174 	 */
175 	ip = mtod(m, struct ip *);
176 	if (m->m_len < iphlen + sizeof(struct udphdr)) {
177 		if ((m = m_pullup(m, iphlen + sizeof(struct udphdr))) == 0) {
178 			udpstat.udps_hdrops++;
179 			return;
180 		}
181 		ip = mtod(m, struct ip *);
182 	}
183 	uh = (struct udphdr *)((caddr_t)ip + iphlen);
184 
185 	/* destination port of 0 is illegal, based on RFC768. */
186 	if (uh->uh_dport == 0)
187 		goto badunlocked;
188 
189 	/*
190 	 * Construct sockaddr format source address.
191 	 * Stuff source address and datagram in user buffer.
192 	 */
193 	bzero(&udp_in, sizeof(udp_in));
194 	udp_in.sin_len = sizeof(udp_in);
195 	udp_in.sin_family = AF_INET;
196 	udp_in.sin_port = uh->uh_sport;
197 	udp_in.sin_addr = ip->ip_src;
198 
199 	/*
200 	 * Make mbuf data length reflect UDP length.
201 	 * If not enough data to reflect UDP length, drop.
202 	 */
203 	len = ntohs((u_short)uh->uh_ulen);
204 	if (ip->ip_len != len) {
205 		if (len > ip->ip_len || len < sizeof(struct udphdr)) {
206 			udpstat.udps_badlen++;
207 			goto badunlocked;
208 		}
209 		m_adj(m, len - ip->ip_len);
210 		/* ip->ip_len = len; */
211 	}
212 	/*
213 	 * Save a copy of the IP header in case we want restore it
214 	 * for sending an ICMP error message in response.
215 	 */
216 	if (!blackhole)
217 		save_ip = *ip;
218 
219 	/*
220 	 * Checksum extended UDP header and data.
221 	 */
222 	if (uh->uh_sum) {
223 		if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
224 			if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
225 				uh->uh_sum = m->m_pkthdr.csum_data;
226 			else
227 				uh->uh_sum = in_pseudo(ip->ip_src.s_addr,
228 				    ip->ip_dst.s_addr, htonl((u_short)len +
229 				    m->m_pkthdr.csum_data + IPPROTO_UDP));
230 			uh->uh_sum ^= 0xffff;
231 		} else {
232 			char b[9];
233 			bcopy(((struct ipovly *)ip)->ih_x1, b, 9);
234 			bzero(((struct ipovly *)ip)->ih_x1, 9);
235 			((struct ipovly *)ip)->ih_len = uh->uh_ulen;
236 			uh->uh_sum = in_cksum(m, len + sizeof (struct ip));
237 			bcopy(b, ((struct ipovly *)ip)->ih_x1, 9);
238 		}
239 		if (uh->uh_sum) {
240 			udpstat.udps_badsum++;
241 			m_freem(m);
242 			return;
243 		}
244 	} else
245 		udpstat.udps_nosum++;
246 
247 	INP_INFO_RLOCK(&udbinfo);
248 
249 	if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
250 	    in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) {
251 		struct inpcb *last;
252 		/*
253 		 * Deliver a multicast or broadcast datagram to *all* sockets
254 		 * for which the local and remote addresses and ports match
255 		 * those of the incoming datagram.  This allows more than
256 		 * one process to receive multi/broadcasts on the same port.
257 		 * (This really ought to be done for unicast datagrams as
258 		 * well, but that would cause problems with existing
259 		 * applications that open both address-specific sockets and
260 		 * a wildcard socket listening to the same port -- they would
261 		 * end up receiving duplicates of every unicast datagram.
262 		 * Those applications open the multiple sockets to overcome an
263 		 * inadequacy of the UDP socket interface, but for backwards
264 		 * compatibility we avoid the problem here rather than
265 		 * fixing the interface.  Maybe 4.5BSD will remedy this?)
266 		 */
267 
268 		/*
269 		 * Locate pcb(s) for datagram.
270 		 * (Algorithm copied from raw_intr().)
271 		 */
272 		last = NULL;
273 		LIST_FOREACH(inp, &udb, inp_list) {
274 			if (inp->inp_lport != uh->uh_dport)
275 				continue;
276 #ifdef INET6
277 			if ((inp->inp_vflag & INP_IPV4) == 0)
278 				continue;
279 #endif
280 			if (inp->inp_laddr.s_addr != INADDR_ANY) {
281 				if (inp->inp_laddr.s_addr != ip->ip_dst.s_addr)
282 					continue;
283 			}
284 			if (inp->inp_faddr.s_addr != INADDR_ANY) {
285 				if (inp->inp_faddr.s_addr !=
286 				    ip->ip_src.s_addr ||
287 				    inp->inp_fport != uh->uh_sport)
288 					continue;
289 			}
290 			INP_LOCK(inp);
291 
292 			/*
293 			 * Check multicast packets to make sure they are only
294 			 * sent to sockets with multicast memberships for the
295 			 * packet's destination address and arrival interface
296 			 */
297 #define MSHIP(_inp, n) ((_inp)->inp_moptions->imo_membership[(n)])
298 #define NMSHIPS(_inp) ((_inp)->inp_moptions->imo_num_memberships)
299 			if (strict_mcast_mship && inp->inp_moptions != NULL) {
300 				int mship, foundmship = 0;
301 
302 				for (mship = 0; mship < NMSHIPS(inp); mship++) {
303 					if (MSHIP(inp, mship)->inm_addr.s_addr
304 					    == ip->ip_dst.s_addr &&
305 					    MSHIP(inp, mship)->inm_ifp
306 					    == m->m_pkthdr.rcvif) {
307 						foundmship = 1;
308 						break;
309 					}
310 				}
311 				if (foundmship == 0) {
312 					INP_UNLOCK(inp);
313 					continue;
314 				}
315 			}
316 #undef NMSHIPS
317 #undef MSHIP
318 			if (last != NULL) {
319 				struct mbuf *n;
320 
321 				n = m_copy(m, 0, M_COPYALL);
322 				if (n != NULL)
323 					udp_append(last, ip, n,
324 						   iphlen +
325 						   sizeof(struct udphdr),
326 						   &udp_in);
327 				INP_UNLOCK(last);
328 			}
329 			last = inp;
330 			/*
331 			 * Don't look for additional matches if this one does
332 			 * not have either the SO_REUSEPORT or SO_REUSEADDR
333 			 * socket options set.  This heuristic avoids searching
334 			 * through all pcbs in the common case of a non-shared
335 			 * port.  It * assumes that an application will never
336 			 * clear these options after setting them.
337 			 */
338 			if ((last->inp_socket->so_options&(SO_REUSEPORT|SO_REUSEADDR)) == 0)
339 				break;
340 		}
341 
342 		if (last == NULL) {
343 			/*
344 			 * No matching pcb found; discard datagram.
345 			 * (No need to send an ICMP Port Unreachable
346 			 * for a broadcast or multicast datgram.)
347 			 */
348 			udpstat.udps_noportbcast++;
349 			goto badheadlocked;
350 		}
351 		udp_append(last, ip, m, iphlen + sizeof(struct udphdr),
352 		    &udp_in);
353 		INP_UNLOCK(last);
354 		INP_INFO_RUNLOCK(&udbinfo);
355 		return;
356 	}
357 	/*
358 	 * Locate pcb for datagram.
359 	 */
360 	inp = in_pcblookup_hash(&udbinfo, ip->ip_src, uh->uh_sport,
361 	    ip->ip_dst, uh->uh_dport, 1, m->m_pkthdr.rcvif);
362 	if (inp == NULL) {
363 		if (log_in_vain) {
364 			char buf[4*sizeof "123"];
365 
366 			strcpy(buf, inet_ntoa(ip->ip_dst));
367 			log(LOG_INFO,
368 			    "Connection attempt to UDP %s:%d from %s:%d\n",
369 			    buf, ntohs(uh->uh_dport), inet_ntoa(ip->ip_src),
370 			    ntohs(uh->uh_sport));
371 		}
372 		udpstat.udps_noport++;
373 		if (m->m_flags & (M_BCAST | M_MCAST)) {
374 			udpstat.udps_noportbcast++;
375 			goto badheadlocked;
376 		}
377 		if (blackhole)
378 			goto badheadlocked;
379 		if (badport_bandlim(BANDLIM_ICMP_UNREACH) < 0)
380 			goto badheadlocked;
381 		*ip = save_ip;
382 		ip->ip_len += iphlen;
383 		icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_PORT, 0, 0);
384 		INP_INFO_RUNLOCK(&udbinfo);
385 		return;
386 	}
387 	INP_LOCK(inp);
388 	/* Check the minimum TTL for socket. */
389 	if (inp->inp_ip_minttl && inp->inp_ip_minttl > ip->ip_ttl)
390 		goto badheadlocked;
391 	udp_append(inp, ip, m, iphlen + sizeof(struct udphdr), &udp_in);
392 	INP_UNLOCK(inp);
393 	INP_INFO_RUNLOCK(&udbinfo);
394 	return;
395 
396 badheadlocked:
397 	if (inp)
398 		INP_UNLOCK(inp);
399 	INP_INFO_RUNLOCK(&udbinfo);
400 badunlocked:
401 	m_freem(m);
402 	if (opts)
403 		m_freem(opts);
404 	return;
405 }
406 
407 /*
408  * Subroutine of udp_input(), which appends the provided mbuf chain to the
409  * passed pcb/socket.  The caller must provide a sockaddr_in via udp_in that
410  * contains the source address.  If the socket ends up being an IPv6 socket,
411  * udp_append() will convert to a sockaddr_in6 before passing the address
412  * into the socket code.
413  */
414 static void
415 udp_append(last, ip, n, off, udp_in)
416 	struct inpcb *last;
417 	struct ip *ip;
418 	struct mbuf *n;
419 	int off;
420 	struct sockaddr_in *udp_in;
421 {
422 	struct sockaddr *append_sa;
423 	struct socket *so;
424 	struct mbuf *opts = 0;
425 #ifdef INET6
426 	struct sockaddr_in6 udp_in6;
427 #endif
428 
429 	INP_LOCK_ASSERT(last);
430 
431 #if defined(IPSEC) || defined(FAST_IPSEC)
432 	/* check AH/ESP integrity. */
433 	if (ipsec4_in_reject(n, last)) {
434 #ifdef IPSEC
435 		ipsecstat.in_polvio++;
436 #endif /*IPSEC*/
437 		m_freem(n);
438 		return;
439 	}
440 #endif /*IPSEC || FAST_IPSEC*/
441 #ifdef MAC
442 	if (mac_check_inpcb_deliver(last, n) != 0) {
443 		m_freem(n);
444 		return;
445 	}
446 #endif
447 	if (last->inp_flags & INP_CONTROLOPTS ||
448 	    last->inp_socket->so_options & (SO_TIMESTAMP | SO_BINTIME)) {
449 #ifdef INET6
450 		if (last->inp_vflag & INP_IPV6) {
451 			int savedflags;
452 
453 			savedflags = last->inp_flags;
454 			last->inp_flags &= ~INP_UNMAPPABLEOPTS;
455 			ip6_savecontrol(last, n, &opts);
456 			last->inp_flags = savedflags;
457 		} else
458 #endif
459 		ip_savecontrol(last, &opts, ip, n);
460 	}
461 #ifdef INET6
462 	if (last->inp_vflag & INP_IPV6) {
463 		bzero(&udp_in6, sizeof(udp_in6));
464 		udp_in6.sin6_len = sizeof(udp_in6);
465 		udp_in6.sin6_family = AF_INET6;
466 		in6_sin_2_v4mapsin6(udp_in, &udp_in6);
467 		append_sa = (struct sockaddr *)&udp_in6;
468 	} else
469 #endif
470 	append_sa = (struct sockaddr *)udp_in;
471 	m_adj(n, off);
472 
473 	so = last->inp_socket;
474 	SOCKBUF_LOCK(&so->so_rcv);
475 	if (sbappendaddr_locked(&so->so_rcv, append_sa, n, opts) == 0) {
476 		m_freem(n);
477 		if (opts)
478 			m_freem(opts);
479 		udpstat.udps_fullsock++;
480 		SOCKBUF_UNLOCK(&so->so_rcv);
481 	} else
482 		sorwakeup_locked(so);
483 }
484 
485 /*
486  * Notify a udp user of an asynchronous error;
487  * just wake up so that he can collect error status.
488  */
489 struct inpcb *
490 udp_notify(inp, errno)
491 	register struct inpcb *inp;
492 	int errno;
493 {
494 	inp->inp_socket->so_error = errno;
495 	sorwakeup(inp->inp_socket);
496 	sowwakeup(inp->inp_socket);
497 	return inp;
498 }
499 
500 void
501 udp_ctlinput(cmd, sa, vip)
502 	int cmd;
503 	struct sockaddr *sa;
504 	void *vip;
505 {
506 	struct ip *ip = vip;
507 	struct udphdr *uh;
508 	struct inpcb *(*notify)(struct inpcb *, int) = udp_notify;
509 	struct in_addr faddr;
510 	struct inpcb *inp;
511 
512 	faddr = ((struct sockaddr_in *)sa)->sin_addr;
513 	if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
514 		return;
515 
516 	/*
517 	 * Redirects don't need to be handled up here.
518 	 */
519 	if (PRC_IS_REDIRECT(cmd))
520 		return;
521 	/*
522 	 * Hostdead is ugly because it goes linearly through all PCBs.
523 	 * XXX: We never get this from ICMP, otherwise it makes an
524 	 * excellent DoS attack on machines with many connections.
525 	 */
526 	if (cmd == PRC_HOSTDEAD)
527 		ip = 0;
528 	else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
529 		return;
530 	if (ip) {
531 		uh = (struct udphdr *)((caddr_t)ip + (ip->ip_hl << 2));
532 		INP_INFO_RLOCK(&udbinfo);
533 		inp = in_pcblookup_hash(&udbinfo, faddr, uh->uh_dport,
534 		    ip->ip_src, uh->uh_sport, 0, NULL);
535 		if (inp != NULL) {
536 			INP_LOCK(inp);
537 			if (inp->inp_socket != NULL) {
538 				(*notify)(inp, inetctlerrmap[cmd]);
539 			}
540 			INP_UNLOCK(inp);
541 		}
542 		INP_INFO_RUNLOCK(&udbinfo);
543 	} else
544 		in_pcbnotifyall(&udbinfo, faddr, inetctlerrmap[cmd], notify);
545 }
546 
547 static int
548 udp_pcblist(SYSCTL_HANDLER_ARGS)
549 {
550 	int error, i, n;
551 	struct inpcb *inp, **inp_list;
552 	inp_gen_t gencnt;
553 	struct xinpgen xig;
554 
555 	/*
556 	 * The process of preparing the TCB list is too time-consuming and
557 	 * resource-intensive to repeat twice on every request.
558 	 */
559 	if (req->oldptr == 0) {
560 		n = udbinfo.ipi_count;
561 		req->oldidx = 2 * (sizeof xig)
562 			+ (n + n/8) * sizeof(struct xinpcb);
563 		return 0;
564 	}
565 
566 	if (req->newptr != 0)
567 		return EPERM;
568 
569 	/*
570 	 * OK, now we're committed to doing something.
571 	 */
572 	INP_INFO_RLOCK(&udbinfo);
573 	gencnt = udbinfo.ipi_gencnt;
574 	n = udbinfo.ipi_count;
575 	INP_INFO_RUNLOCK(&udbinfo);
576 
577 	error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
578 		+ n * sizeof(struct xinpcb));
579 	if (error != 0)
580 		return (error);
581 
582 	xig.xig_len = sizeof xig;
583 	xig.xig_count = n;
584 	xig.xig_gen = gencnt;
585 	xig.xig_sogen = so_gencnt;
586 	error = SYSCTL_OUT(req, &xig, sizeof xig);
587 	if (error)
588 		return error;
589 
590 	inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
591 	if (inp_list == 0)
592 		return ENOMEM;
593 
594 	INP_INFO_RLOCK(&udbinfo);
595 	for (inp = LIST_FIRST(udbinfo.listhead), i = 0; inp && i < n;
596 	     inp = LIST_NEXT(inp, inp_list)) {
597 		INP_LOCK(inp);
598 		if (inp->inp_gencnt <= gencnt &&
599 		    cr_canseesocket(req->td->td_ucred, inp->inp_socket) == 0)
600 			inp_list[i++] = inp;
601 		INP_UNLOCK(inp);
602 	}
603 	INP_INFO_RUNLOCK(&udbinfo);
604 	n = i;
605 
606 	error = 0;
607 	for (i = 0; i < n; i++) {
608 		inp = inp_list[i];
609 		if (inp->inp_gencnt <= gencnt) {
610 			struct xinpcb xi;
611 			bzero(&xi, sizeof(xi));
612 			xi.xi_len = sizeof xi;
613 			/* XXX should avoid extra copy */
614 			bcopy(inp, &xi.xi_inp, sizeof *inp);
615 			if (inp->inp_socket)
616 				sotoxsocket(inp->inp_socket, &xi.xi_socket);
617 			xi.xi_inp.inp_gencnt = inp->inp_gencnt;
618 			error = SYSCTL_OUT(req, &xi, sizeof xi);
619 		}
620 	}
621 	if (!error) {
622 		/*
623 		 * Give the user an updated idea of our state.
624 		 * If the generation differs from what we told
625 		 * her before, she knows that something happened
626 		 * while we were processing this request, and it
627 		 * might be necessary to retry.
628 		 */
629 		INP_INFO_RLOCK(&udbinfo);
630 		xig.xig_gen = udbinfo.ipi_gencnt;
631 		xig.xig_sogen = so_gencnt;
632 		xig.xig_count = udbinfo.ipi_count;
633 		INP_INFO_RUNLOCK(&udbinfo);
634 		error = SYSCTL_OUT(req, &xig, sizeof xig);
635 	}
636 	free(inp_list, M_TEMP);
637 	return error;
638 }
639 
640 SYSCTL_PROC(_net_inet_udp, UDPCTL_PCBLIST, pcblist, CTLFLAG_RD, 0, 0,
641 	    udp_pcblist, "S,xinpcb", "List of active UDP sockets");
642 
643 static int
644 udp_getcred(SYSCTL_HANDLER_ARGS)
645 {
646 	struct xucred xuc;
647 	struct sockaddr_in addrs[2];
648 	struct inpcb *inp;
649 	int error;
650 
651 	error = suser_cred(req->td->td_ucred, SUSER_ALLOWJAIL);
652 	if (error)
653 		return (error);
654 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
655 	if (error)
656 		return (error);
657 	INP_INFO_RLOCK(&udbinfo);
658 	inp = in_pcblookup_hash(&udbinfo, addrs[1].sin_addr, addrs[1].sin_port,
659 				addrs[0].sin_addr, addrs[0].sin_port, 1, NULL);
660 	if (inp == NULL || inp->inp_socket == NULL) {
661 		error = ENOENT;
662 		goto out;
663 	}
664 	error = cr_canseesocket(req->td->td_ucred, inp->inp_socket);
665 	if (error)
666 		goto out;
667 	cru2x(inp->inp_socket->so_cred, &xuc);
668 out:
669 	INP_INFO_RUNLOCK(&udbinfo);
670 	if (error == 0)
671 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
672 	return (error);
673 }
674 
675 SYSCTL_PROC(_net_inet_udp, OID_AUTO, getcred,
676     CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
677     udp_getcred, "S,xucred", "Get the xucred of a UDP connection");
678 
679 static int
680 udp_output(inp, m, addr, control, td)
681 	register struct inpcb *inp;
682 	struct mbuf *m;
683 	struct sockaddr *addr;
684 	struct mbuf *control;
685 	struct thread *td;
686 {
687 	register struct udpiphdr *ui;
688 	register int len = m->m_pkthdr.len;
689 	struct in_addr faddr, laddr;
690 	struct cmsghdr *cm;
691 	struct sockaddr_in *sin, src;
692 	int error = 0;
693 	int ipflags;
694 	u_short fport, lport;
695 	int unlock_udbinfo;
696 
697 	/*
698 	 * udp_output() may need to temporarily bind or connect the current
699 	 * inpcb.  As such, we don't know up front what inpcb locks we will
700 	 * need.  Do any work to decide what is needed up front before
701 	 * acquiring locks.
702 	 */
703 	if (len + sizeof(struct udpiphdr) > IP_MAXPACKET) {
704 		if (control)
705 			m_freem(control);
706 		m_freem(m);
707 		return EMSGSIZE;
708 	}
709 
710 	src.sin_addr.s_addr = INADDR_ANY;
711 	if (control != NULL) {
712 		/*
713 		 * XXX: Currently, we assume all the optional information
714 		 * is stored in a single mbuf.
715 		 */
716 		if (control->m_next) {
717 			m_freem(control);
718 			m_freem(m);
719 			return EINVAL;
720 		}
721 		for (; control->m_len > 0;
722 		    control->m_data += CMSG_ALIGN(cm->cmsg_len),
723 		    control->m_len -= CMSG_ALIGN(cm->cmsg_len)) {
724 			cm = mtod(control, struct cmsghdr *);
725 			if (control->m_len < sizeof(*cm) || cm->cmsg_len == 0 ||
726 			    cm->cmsg_len > control->m_len) {
727 				error = EINVAL;
728 				break;
729 			}
730 			if (cm->cmsg_level != IPPROTO_IP)
731 				continue;
732 
733 			switch (cm->cmsg_type) {
734 			case IP_SENDSRCADDR:
735 				if (cm->cmsg_len !=
736 				    CMSG_LEN(sizeof(struct in_addr))) {
737 					error = EINVAL;
738 					break;
739 				}
740 				bzero(&src, sizeof(src));
741 				src.sin_family = AF_INET;
742 				src.sin_len = sizeof(src);
743 				src.sin_port = inp->inp_lport;
744 				src.sin_addr = *(struct in_addr *)CMSG_DATA(cm);
745 				break;
746 			default:
747 				error = ENOPROTOOPT;
748 				break;
749 			}
750 			if (error)
751 				break;
752 		}
753 		m_freem(control);
754 	}
755 	if (error) {
756 		m_freem(m);
757 		return error;
758 	}
759 
760 	if (src.sin_addr.s_addr != INADDR_ANY ||
761 	    addr != NULL) {
762 		INP_INFO_WLOCK(&udbinfo);
763 		unlock_udbinfo = 1;
764 	} else
765 		unlock_udbinfo = 0;
766 	INP_LOCK(inp);
767 
768 #ifdef MAC
769 	mac_create_mbuf_from_inpcb(inp, m);
770 #endif
771 
772 	laddr = inp->inp_laddr;
773 	lport = inp->inp_lport;
774 	if (src.sin_addr.s_addr != INADDR_ANY) {
775 		if (lport == 0) {
776 			error = EINVAL;
777 			goto release;
778 		}
779 		error = in_pcbbind_setup(inp, (struct sockaddr *)&src,
780 		    &laddr.s_addr, &lport, td->td_ucred);
781 		if (error)
782 			goto release;
783 	}
784 
785 	if (addr) {
786 		sin = (struct sockaddr_in *)addr;
787 		if (jailed(td->td_ucred))
788 			prison_remote_ip(td->td_ucred, 0, &sin->sin_addr.s_addr);
789 		if (inp->inp_faddr.s_addr != INADDR_ANY) {
790 			error = EISCONN;
791 			goto release;
792 		}
793 		error = in_pcbconnect_setup(inp, addr, &laddr.s_addr, &lport,
794 		    &faddr.s_addr, &fport, NULL, td->td_ucred);
795 		if (error)
796 			goto release;
797 
798 		/* Commit the local port if newly assigned. */
799 		if (inp->inp_laddr.s_addr == INADDR_ANY &&
800 		    inp->inp_lport == 0) {
801 			/*
802 			 * Remember addr if jailed, to prevent rebinding.
803 			 */
804 			if (jailed(td->td_ucred))
805 				inp->inp_laddr = laddr;
806 			inp->inp_lport = lport;
807 			if (in_pcbinshash(inp) != 0) {
808 				inp->inp_lport = 0;
809 				error = EAGAIN;
810 				goto release;
811 			}
812 			inp->inp_flags |= INP_ANONPORT;
813 		}
814 	} else {
815 		faddr = inp->inp_faddr;
816 		fport = inp->inp_fport;
817 		if (faddr.s_addr == INADDR_ANY) {
818 			error = ENOTCONN;
819 			goto release;
820 		}
821 	}
822 
823 	/*
824 	 * Calculate data length and get a mbuf for UDP, IP, and possible
825 	 * link-layer headers.  Immediate slide the data pointer back forward
826 	 * since we won't use that space at this layer.
827 	 */
828 	M_PREPEND(m, sizeof(struct udpiphdr) + max_linkhdr, M_DONTWAIT);
829 	if (m == NULL) {
830 		error = ENOBUFS;
831 		goto release;
832 	}
833 	m->m_data += max_linkhdr;
834 	m->m_len -= max_linkhdr;
835 	m->m_pkthdr.len -= max_linkhdr;
836 
837 	/*
838 	 * Fill in mbuf with extended UDP header
839 	 * and addresses and length put into network format.
840 	 */
841 	ui = mtod(m, struct udpiphdr *);
842 	bzero(ui->ui_x1, sizeof(ui->ui_x1));	/* XXX still needed? */
843 	ui->ui_pr = IPPROTO_UDP;
844 	ui->ui_src = laddr;
845 	ui->ui_dst = faddr;
846 	ui->ui_sport = lport;
847 	ui->ui_dport = fport;
848 	ui->ui_ulen = htons((u_short)len + sizeof(struct udphdr));
849 
850 	/*
851 	 * Set the Don't Fragment bit in the IP header.
852 	 */
853 	if (inp->inp_flags & INP_DONTFRAG) {
854 		struct ip *ip;
855 		ip = (struct ip *)&ui->ui_i;
856 		ip->ip_off |= IP_DF;
857 	}
858 
859 	ipflags = 0;
860 	if (inp->inp_socket->so_options & SO_DONTROUTE)
861 		ipflags |= IP_ROUTETOIF;
862 	if (inp->inp_socket->so_options & SO_BROADCAST)
863 		ipflags |= IP_ALLOWBROADCAST;
864 	if (inp->inp_vflag & INP_ONESBCAST)
865 		ipflags |= IP_SENDONES;
866 
867 	/*
868 	 * Set up checksum and output datagram.
869 	 */
870 	if (udpcksum) {
871 		if (inp->inp_vflag & INP_ONESBCAST)
872 			faddr.s_addr = INADDR_BROADCAST;
873 		ui->ui_sum = in_pseudo(ui->ui_src.s_addr, faddr.s_addr,
874 		    htons((u_short)len + sizeof(struct udphdr) + IPPROTO_UDP));
875 		m->m_pkthdr.csum_flags = CSUM_UDP;
876 		m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum);
877 	} else {
878 		ui->ui_sum = 0;
879 	}
880 	((struct ip *)ui)->ip_len = sizeof (struct udpiphdr) + len;
881 	((struct ip *)ui)->ip_ttl = inp->inp_ip_ttl;	/* XXX */
882 	((struct ip *)ui)->ip_tos = inp->inp_ip_tos;	/* XXX */
883 	udpstat.udps_opackets++;
884 
885 	if (unlock_udbinfo)
886 		INP_INFO_WUNLOCK(&udbinfo);
887 	error = ip_output(m, inp->inp_options, NULL, ipflags,
888 	    inp->inp_moptions, inp);
889 	INP_UNLOCK(inp);
890 	return (error);
891 
892 release:
893 	INP_UNLOCK(inp);
894 	if (unlock_udbinfo)
895 		INP_INFO_WUNLOCK(&udbinfo);
896 	m_freem(m);
897 	return (error);
898 }
899 
900 u_long	udp_sendspace = 9216;		/* really max datagram size */
901 					/* 40 1K datagrams */
902 SYSCTL_INT(_net_inet_udp, UDPCTL_MAXDGRAM, maxdgram, CTLFLAG_RW,
903     &udp_sendspace, 0, "Maximum outgoing UDP datagram size");
904 
905 u_long	udp_recvspace = 40 * (1024 +
906 #ifdef INET6
907 				      sizeof(struct sockaddr_in6)
908 #else
909 				      sizeof(struct sockaddr_in)
910 #endif
911 				      );
912 SYSCTL_INT(_net_inet_udp, UDPCTL_RECVSPACE, recvspace, CTLFLAG_RW,
913     &udp_recvspace, 0, "Maximum space for incoming UDP datagrams");
914 
915 static int
916 udp_abort(struct socket *so)
917 {
918 	struct inpcb *inp;
919 
920 	INP_INFO_WLOCK(&udbinfo);
921 	inp = sotoinpcb(so);
922 	if (inp == 0) {
923 		INP_INFO_WUNLOCK(&udbinfo);
924 		return EINVAL;	/* ??? possible? panic instead? */
925 	}
926 	INP_LOCK(inp);
927 	soisdisconnected(so);
928 	in_pcbdetach(inp);
929 	INP_INFO_WUNLOCK(&udbinfo);
930 	return 0;
931 }
932 
933 static int
934 udp_attach(struct socket *so, int proto, struct thread *td)
935 {
936 	struct inpcb *inp;
937 	int error;
938 
939 	INP_INFO_WLOCK(&udbinfo);
940 	inp = sotoinpcb(so);
941 	if (inp != 0) {
942 		INP_INFO_WUNLOCK(&udbinfo);
943 		return EINVAL;
944 	}
945 	error = soreserve(so, udp_sendspace, udp_recvspace);
946 	if (error) {
947 		INP_INFO_WUNLOCK(&udbinfo);
948 		return error;
949 	}
950 	error = in_pcballoc(so, &udbinfo, "udpinp");
951 	if (error) {
952 		INP_INFO_WUNLOCK(&udbinfo);
953 		return error;
954 	}
955 
956 	inp = (struct inpcb *)so->so_pcb;
957 	INP_LOCK(inp);
958 	INP_INFO_WUNLOCK(&udbinfo);
959 	inp->inp_vflag |= INP_IPV4;
960 	inp->inp_ip_ttl = ip_defttl;
961 	INP_UNLOCK(inp);
962 	return 0;
963 }
964 
965 static int
966 udp_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
967 {
968 	struct inpcb *inp;
969 	int error;
970 
971 	INP_INFO_WLOCK(&udbinfo);
972 	inp = sotoinpcb(so);
973 	if (inp == 0) {
974 		INP_INFO_WUNLOCK(&udbinfo);
975 		return EINVAL;
976 	}
977 	INP_LOCK(inp);
978 	error = in_pcbbind(inp, nam, td->td_ucred);
979 	INP_UNLOCK(inp);
980 	INP_INFO_WUNLOCK(&udbinfo);
981 	return error;
982 }
983 
984 static int
985 udp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
986 {
987 	struct inpcb *inp;
988 	int error;
989 	struct sockaddr_in *sin;
990 
991 	INP_INFO_WLOCK(&udbinfo);
992 	inp = sotoinpcb(so);
993 	if (inp == 0) {
994 		INP_INFO_WUNLOCK(&udbinfo);
995 		return EINVAL;
996 	}
997 	INP_LOCK(inp);
998 	if (inp->inp_faddr.s_addr != INADDR_ANY) {
999 		INP_UNLOCK(inp);
1000 		INP_INFO_WUNLOCK(&udbinfo);
1001 		return EISCONN;
1002 	}
1003 	sin = (struct sockaddr_in *)nam;
1004 	if (jailed(td->td_ucred))
1005 		prison_remote_ip(td->td_ucred, 0, &sin->sin_addr.s_addr);
1006 	error = in_pcbconnect(inp, nam, td->td_ucred);
1007 	if (error == 0)
1008 		soisconnected(so);
1009 	INP_UNLOCK(inp);
1010 	INP_INFO_WUNLOCK(&udbinfo);
1011 	return error;
1012 }
1013 
1014 static int
1015 udp_detach(struct socket *so)
1016 {
1017 	struct inpcb *inp;
1018 
1019 	INP_INFO_WLOCK(&udbinfo);
1020 	inp = sotoinpcb(so);
1021 	if (inp == 0) {
1022 		INP_INFO_WUNLOCK(&udbinfo);
1023 		return EINVAL;
1024 	}
1025 	INP_LOCK(inp);
1026 	in_pcbdetach(inp);
1027 	INP_INFO_WUNLOCK(&udbinfo);
1028 	return 0;
1029 }
1030 
1031 static int
1032 udp_disconnect(struct socket *so)
1033 {
1034 	struct inpcb *inp;
1035 
1036 	INP_INFO_WLOCK(&udbinfo);
1037 	inp = sotoinpcb(so);
1038 	if (inp == 0) {
1039 		INP_INFO_WUNLOCK(&udbinfo);
1040 		return EINVAL;
1041 	}
1042 	INP_LOCK(inp);
1043 	if (inp->inp_faddr.s_addr == INADDR_ANY) {
1044 		INP_INFO_WUNLOCK(&udbinfo);
1045 		INP_UNLOCK(inp);
1046 		return ENOTCONN;
1047 	}
1048 
1049 	in_pcbdisconnect(inp);
1050 	inp->inp_laddr.s_addr = INADDR_ANY;
1051 	INP_UNLOCK(inp);
1052 	INP_INFO_WUNLOCK(&udbinfo);
1053 	so->so_state &= ~SS_ISCONNECTED;		/* XXX */
1054 	return 0;
1055 }
1056 
1057 static int
1058 udp_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *addr,
1059 	    struct mbuf *control, struct thread *td)
1060 {
1061 	struct inpcb *inp;
1062 
1063 	inp = sotoinpcb(so);
1064 	return udp_output(inp, m, addr, control, td);
1065 }
1066 
1067 int
1068 udp_shutdown(struct socket *so)
1069 {
1070 	struct inpcb *inp;
1071 
1072 	INP_INFO_RLOCK(&udbinfo);
1073 	inp = sotoinpcb(so);
1074 	if (inp == 0) {
1075 		INP_INFO_RUNLOCK(&udbinfo);
1076 		return EINVAL;
1077 	}
1078 	INP_LOCK(inp);
1079 	INP_INFO_RUNLOCK(&udbinfo);
1080 	socantsendmore(so);
1081 	INP_UNLOCK(inp);
1082 	return 0;
1083 }
1084 
1085 /*
1086  * This is the wrapper function for in_setsockaddr.  We just pass down
1087  * the pcbinfo for in_setsockaddr to lock.  We don't want to do the locking
1088  * here because in_setsockaddr will call malloc and might block.
1089  */
1090 static int
1091 udp_sockaddr(struct socket *so, struct sockaddr **nam)
1092 {
1093 	return (in_setsockaddr(so, nam, &udbinfo));
1094 }
1095 
1096 /*
1097  * This is the wrapper function for in_setpeeraddr.  We just pass down
1098  * the pcbinfo for in_setpeeraddr to lock.
1099  */
1100 static int
1101 udp_peeraddr(struct socket *so, struct sockaddr **nam)
1102 {
1103 	return (in_setpeeraddr(so, nam, &udbinfo));
1104 }
1105 
1106 struct pr_usrreqs udp_usrreqs = {
1107 	.pru_abort =		udp_abort,
1108 	.pru_attach =		udp_attach,
1109 	.pru_bind =		udp_bind,
1110 	.pru_connect =		udp_connect,
1111 	.pru_control =		in_control,
1112 	.pru_detach =		udp_detach,
1113 	.pru_disconnect =	udp_disconnect,
1114 	.pru_peeraddr =		udp_peeraddr,
1115 	.pru_send =		udp_send,
1116 	.pru_shutdown =		udp_shutdown,
1117 	.pru_sockaddr =		udp_sockaddr,
1118 	.pru_sosetlabel =	in_pcbsosetlabel
1119 };
1120